tradein-mvp/backend/app/services/yandex_address_backfill.py

@@ -26,15 +26,20 @@ from sqlalchemy.orm import Session
 
 logger = logging.getLogger(__name__)
 
-# Extract 'Екатеринбург, <full address>' from <title> — matches:
+# Extract '<City>, <full address>' from <title> — matches any city, not just Ekaterinburg.
+# Handles EKB metro-area satellite towns: Верхняя Пышма, Берёзовский, Среднеуральск, Арамиль, etc.
+# Examples:
 #   «Продажа квартиры — Екатеринбург, улица Горького, 36 — id 7654321 на Яндекс.Недвижимости»
-# Group 1 = «Екатеринбург, улица Горького, 36» (stripped of trailing punctuation/spaces)
+#   «Купить квартиру … — ЖК «Дуэт», Верхняя Пышма, улица Мальцева, 14 — id 75332…»
+#   «Купить квартиру … — ЖК «Уют-Сити», Берёзовский, Александровский проспект, 3 — id 3988…»
+# Group 1 = «<City>, <street/house>» (stripped of trailing punctuation/spaces)
+# The optional ЖК-prefix (?:[^—,]*?,\s*) consumes «ЖК «Дуэт», » before the city name.
+# City capture starts with [А-ЯЁ] to ensure the prefix never swallows the city itself.
 _RE_TITLE_ADDRESS = re.compile(
     r"<title>[^<]*?—\s*"
-    r"(?:[^—,]*?,\s*)?"  # optional ЖК/жилой комплекс prefix
-    r"(Екатеринбург,\s*[^<—]+?)"
+    r"(?:(?![А-ЯЁ][а-яё])[^—,]*?,\s*)?"  # optional ЖК-prefix — fires only when NOT a city name
+    r"([А-ЯЁ][^—<]+?,\s*[^<—]+?)"  # <City>, <street/house> — any city (uppercase Cyrillic start)
     r"\s*—\s*id\s*\d+[^<]*</title>",  # allow trailing text after id NNN (e.g. 'на Яндекс...')
-    re.IGNORECASE,
 )
 
 # Listings without a house number: address contains no digit after comma-space
@@ -53,7 +58,11 @@ class YandexAddressBackfillResult:
 
 
 def _extract_address_from_title(html: str) -> str | None:
-    """Extract full address from Yandex offer detail page <title>."""
+    """Extract full address from Yandex offer detail page <title>.
+
+    City-agnostic: works for EKB and satellite towns (Верхняя Пышма, Берёзовский, etc.).
+    Returns «<City>, <street>, <house>» or None if the title format is not recognised.
+    """
     html_clean = html.replace("\xa0", " ")
     m = _RE_TITLE_ADDRESS.search(html_clean)
     if not m:

tradein-mvp/backend/tests/tasks/test_yandex_address_backfill.py

@@ -60,6 +60,67 @@ def test_extract_address_with_zhk_prefix() -> None:
     assert "Екатеринбург" in addr
 
 
+# ---------------------------------------------------------------------------
+# City-agnostic tests — satellite towns (T10 live-bug fix)
+# ---------------------------------------------------------------------------
+
+
+def test_extract_address_verkhnaya_pyshma_with_zhk() -> None:
+    """Real title from sweep #28: ЖК «Дуэт», Верхняя Пышма — was returning None before fix."""
+    html = (
+        "<title>Купить квартиру 102.4 м²"
+        " — ЖК «Дуэт», Верхняя Пышма, улица Мальцева, 14"
+        " — id 7533242317391171630</title>"
+    )
+    addr = _extract_address_from_title(html)
+    assert addr is not None, "Верхняя Пышма title should produce an address"
+    assert _RE_HAS_HOUSE_NUMBER.search(addr), f"Should contain house number: {addr!r}"
+    assert "Верхняя Пышма" in addr
+    assert "Мальцева" in addr
+    assert "14" in addr
+
+
+def test_extract_address_beryozovsky_with_zhk() -> None:
+    """Real title from sweep #28: ЖК «Уют-Сити», Берёзовский — was returning None before fix."""
+    html = (
+        "<title>Купить квартиру 139.2 м²"
+        " — ЖК «Уют-Сити», Берёзовский, Александровский проспект, 3"
+        " — id 3988368535316036933</title>"
+    )
+    addr = _extract_address_from_title(html)
+    assert addr is not None, "Берёзовский title should produce an address"
+    assert _RE_HAS_HOUSE_NUMBER.search(addr), f"Should contain house number: {addr!r}"
+    assert "Берёзовский" in addr
+    assert "Александровский" in addr
+    assert "3" in addr
+
+
+def test_extract_address_ekb_no_zhk_still_matches() -> None:
+    """EKB variant from docstring (no ЖК prefix) still matches after city-agnostic change."""
+    html = "<title>Продажа квартиры — Екатеринбург, улица Горького, 36" " — id 7654321</title>"
+    addr = _extract_address_from_title(html)
+    assert addr is not None, "EKB no-ЖК title should still match"
+    assert _RE_HAS_HOUSE_NUMBER.search(addr), f"Should contain house number: {addr!r}"
+    assert "Екатеринбург" in addr
+    assert "Горького" in addr
+    assert "36" in addr
+
+
+def test_extract_address_no_house_number_guard() -> None:
+    """Street-only title (no house number): extractor may return value, but guard rejects it.
+
+    _RE_HAS_HOUSE_NUMBER must NOT match → backfill skips the listing to avoid overwriting
+    with another street-only address.
+    """
+    html = "<title>Продажа квартиры — Екатеринбург, улица Мира — id 999888</title>"
+    addr = _extract_address_from_title(html)
+    # If the regex matches at all, the extracted value must not contain a house number.
+    if addr is not None:
+        assert not _RE_HAS_HOUSE_NUMBER.search(
+            addr
+        ), f"Street-only addr must fail house-number guard: {addr!r}"
+
+
 def test_extract_address_nbsp_replaced() -> None:
     """\\xa0 (non-breaking space) in title should be normalised before regex match."""
     html = (
@@ -87,22 +148,6 @@ def test_extract_address_title_without_id_returns_none() -> None:
     assert addr is None
 
 
-def test_extract_address_street_only_still_extracted() -> None:
-    """Title with street-only (no house number) is extracted but fails the house-number check.
-
-    The extractor itself does not gate on house number presence — the caller
-    (backfill_yandex_addresses) checks _RE_HAS_HOUSE_NUMBER separately and skips.
-    """
-    html = "<title>Продажа квартиры — Екатеринбург, улица Горького — id 7654321</title>"
-    addr = _extract_address_from_title(html)
-    # May or may not match depending on whether «улица Горького» satisfies the regex —
-    # the key assertion is that IF it does, it does NOT contain a house number.
-    if addr is not None:
-        assert not _RE_HAS_HOUSE_NUMBER.search(
-            addr
-        ), "Street-only extracted address should NOT contain a house number"
-
-
 # ---------------------------------------------------------------------------
 # Task-level smoke tests (no real DB, no real HTTP)
 # ---------------------------------------------------------------------------